Recently, the 6th World Conference on Artificial Intelligence (WAIC) was successfully held in Shanghai, and the attention of this conference is unprecedented. It is worth mentioning that in addition to the general large model displayed by the Internet giant, many enterprises focusing on vertical fields have also launched their own vertical large models. For example, Ronglian Cloud launched the big language model "Chitu Big Model"; Daguan Data released a large model of "Cao Zhi" with long text, vertical and multilingual characteristics; Mido officially released the first vertical large language model created for the field of intelligent proofreading in China - "Mido Wenxiu".
Tang Daosheng, CEO of Tencent Cloud and Smart Industry Business Group, pointed out at the conference: "The general big model has strong capabilities, but it cannot solve the specific problems of many enterprises. Building your own exclusive model based on the large model of the industry may be a better choice for enterprises. He believes that with the development of the big language model, industry and society will move from digitalization and networking to intelligence.
According to incomplete statistics, more than 30 large models were unveiled this year's WAIC, promoting artificial intelligence from perception to cognition, from recognition to generation, and from general to specialized from different dimensions. So, when large models are so popular, how to predict the development trend of large models? The "Generative AI" report released at the conference believes that China will have a multilingual general large model comparable to GPT-4, and the current generative AI market is in the early stage of technology dominance, and there are opportunities for platform companies with a market value of hundreds of billions of dollars.
At the bottom of the larger model, computing power and chips play a key role. At this year's WAIC, Huawei and its partners released the Ascend AI large-model training and promotion integrated solution, accelerating the application of large models in various industries. As a domestic strategic partner of the "Cao Zhi" large model, the cloud service vendor UCloud AIGC computing power base was also unveiled at the conference. In addition, as a representative manufacturer of domestic GPGPU, Zhixin exhibited the general-purpose GPU "Tianyao 100" and the inference product "Zhikai 100".
At this conference, a number of vertical large models were unveiled, what functions do they have? Xiao Yanghua, director of Shanghai Key Laboratory of Data Science and professor at Fudan University, said: "Big models are not 'fireworks shows', but to become advanced productivity, and this wave of AGI revolution is a new 'meta-revolution'. The big model is of great significance to the development of the industry, and with the big model, enterprises have ushered in a new opportunity to achieve an 'end-to-end' value monetization. ”
Chen Yunwen, chairman and CEO of Daguan Data, released the "Cao Zhi" large model at the forum and officially opened it for public testing. According to reports, the "Cao Zhi" large model has the characteristics of long text, verticalization and multi-language, and supports automatic writing and multilingual translation of long text in multiple languages by connecting multiple models and knowledge graphs in parallel, and comprehensively empowers scenarios such as long document writing, machine translation, semantic analysis review, knowledge question answering, text-to-SQL, etc., and can be deeply optimized and personalized for the copywriting needs of different industries and fields.
Liu Jie, President of Ronglian Cloud AI Research Institute, delivered a keynote speech at the forum, introducing the professionalism and commercialization of the "Red Rabbit Model". He said that the "Chitu Big Model" is a multi-level large-language model for vertical industries for enterprise applications, which empowers the communication intelligence 2.0 era by reconstructing the value of intelligent customer service and digital intelligent marketing. Based on the "Chitu Model", enterprises can build their own intelligent customer service and digital intelligent marketing, and complete the evolution from "cost reduction and efficiency increase" to "value creation".
Zhai Guangjing, CEO of Midu, officially released the first vertical large-language model built for the field of intelligent proofreading in China - Midu Wenxiu at the conference. Based on the Big Language Model (LLM), Midu uses high-quality data to learn a variety of characteristic sub-tasks, greatly improving the intelligence of Chinese proofreading and editing capabilities. iFLYTEK and Midu signed a cooperation agreement at the conference site, and in the future, the two parties will deepen their respective technical advantages, jointly promote the advancement of the intelligent proofreading industry, jointly empower the industry ecology, and provide high-quality intelligent proofreading services for more enterprises and institutions.
Computing power is the key to the development of large models. Without large computing power, making a large model is a fantasy. Ken Hu, Huawei's rotating chairman, said that on the basis of each single point of innovation, Huawei has given full play to the comprehensive advantages of cloud, computing, storage, networking, and energy, carried out architectural innovation, and launched the Ascend AI cluster, which is equivalent to designing the AI data center as a supercomputer. At present, the Ascend AI cluster solution has been applied in artificial intelligence computing centers and key laboratories in 25 cities across the country, of which 7, including Wuhan and Xi'an, have become the first batch of "National New Generation Artificial Intelligence Public Computing Power Open Innovation Platform" recognized by the Ministry of Science and Technology.
Under NVIDIA's strong market position, the landing problem of domestic GPUs has always been a challenge that plagues the improvement of domestic AI computing power. However, with a certain amount of technology accumulation, a number of start-up domestic GPU manufacturers have begun to emerge. In the past two years, Tianyuan Zhixin has successively released the general GPU training product "Tianyao 100" and the inference product "Zhiqiao 100", and after the adaptation verification of a number of partners, it has achieved remarkable results in the application landing level, especially in the field of large models, the company built a 40P computing power 320 Tianyao 100 acceleration card computing power cluster in the first half of the year, completed the full training of KLCII's 7 billion parameter large model, and is currently the only domestic general-purpose GPU product that can support the complete training of large models.
At this year's WAIC, Days Zhixin demonstrated the remarkable progress made in large model training and reasoning, among which the exhibition points such as image recognition/image search, 3D modeling, and large model reasoning attracted many attention. A variety of application scenarios such as smart retail, smart OCR, intelligent computing center, and object detection exhibited at the same time also attracted visitors to stop and watch.
In this article, we can see the application and development of large models in various fields. The big model can not only help enterprises solve practical problems, but also promote the industry and society from digitalization and networking to intelligence. With the development of large models, we have reason to believe that artificial intelligence will play a greater role in various fields in the future, bringing more convenience and surprises to our lives. At the same time, we should also see that the development of large models is inseparable from the support of strong computing power. In this regard, domestic GPU manufacturers have made significant progress, providing strong support for the development of large models. In short, the development prospects of large models are worth looking forward to, and we have reason to believe that it will bring more opportunities and challenges to the field of artificial intelligence.